During index building, elasticlunr-rs iterates over the token &str's content in Unicode Scalar Values.

While the JS library does it in this way:

elasticlunr.InvertedIndex.prototype.addToken = function (token, tokenInfo, root) {
  var root = root || this.root,
      idx = 0;

  while (idx <= token.length - 1) {
    var key = token[idx];

The JS string is actually iterated in UTF-16 Code Units, which are entire characters for English, most alphabetic text, common Chinese characters; but not Emojis and rare Chinese characters.

Related issue with mdBook.